The Software Is the Cheap Line

Founders send us a link to a brokerage app and ask what it costs to build. The honest 2026 answer is a range so wide it looks like nobody did the work. They did. They scoped different products and then used the same noun.

One experienced-freelancer ledger for a lean loop — browse a symbol, show a live price, place an order, view a portfolio — lands around $11,000 to $23,000 and 8 to 14 weeks, software only. Agency guides published the same year put a trading or paper-trading dashboard closer to €40,000–€120,000, or $40,000 to $120,000, once charts, an order-management sketch, and backtesting are in the sentence. A wealth-style MVP with paper trading, watchlists, and a read-only portfolio import shows up around $40,000 to $60,000 in one breakdown, with live routing and tax lots pushing past $90,000. Payments MVPs sit cheaper, often in the mid-teens of thousands, because a card processor already owns the compliance you would otherwise build.

The spread is not a scam. It is the difference between a paper ledger and a broker. We will not pretend those are one sprint.

What the software number leaves out

Every serious 2026 cost piece we have read says the same second sentence. The app is not the bill.

Licensed market data is the line founders discover after the proposal. Real-time US quotes are a contract with redistribution rules, not a JSON endpoint you found in a README. Retail-scale display is often budgeted at roughly $500 to $5,000 a month, and it scales with users. Delayed data is a different, usually cheaper, contract. It is still a contract. Paper trading does not waive it. If the prices on the phone are real, someone is paying a display fee whether or not a share ever changes hands. We write the vendor, the delay, and the redistribution answer into the work order. "We will use an API" is not an answer.

Brokerage-as-a-service, KYC per check, and counsel are the next layer, and several guides say it exceeds the software the moment cash moves. A SOC 2 conversation or a money-transmitter question belongs to that world. It does not belong to a paper ledger. Mixing them into version one is how a three-week sprint becomes a year of waiting on people who do not work at our desk.

Compliance regimes change the shape of the build more than the screen count. A wallet that tokenizes cards through a processor is a different product from anything that routes an order. We are not your counsel. We will not freelance which registration applies. We will refuse a fixed price on "figure out the license and also build the app." Discovery can list the questions. A lawyer answers them. The sprint builds only what those answers allow.

Why fixed price only works after the noun is specific

Hourly billing hides this. The clock runs while everyone argues about whether "trading" includes a blotter. A fixed-price sprint cannot. Our work orders name a deliverable that fits in two or three weeks: a paper ledger with double-entry cash and positions, one named data contract, a watchlist, a staging URL, and a reconciliation test. Charts beyond a single price series, options, fractional shares, and anything that routes to a venue are different orders. If they appear mid-sprint, we stop and write them down. We do not quietly eat them.

That is also why public price bands are marketing until the data contract is chosen. A team that says "$20,000 for a Robinhood" without a feed, a ledger definition, and a statement of what is paper versus live is selling a slide. A team that says "three weeks, this feed, this ledger, this staging URL, this price" is selling a sprint. We only do the second one. Two seniors, no handoff, a definition of done you can account.

Discovery is the cheap way to get to that sentence. Public guides put a one-to-two-week technical discovery in the low thousands of euros or dollars, often credited toward the build. We use it to kill scope, not to produce a pitch deck. The output is a risk register: data licensing, session calendar, corporate actions, what a split does to the paper book, and which regulations the counsel still has to read. If discovery cannot name those, the build estimate is fiction and we will not sign it.

Rehearse before you buy the feed

You can learn whether the product is a watchlist with a paper ticket, or a brokerage, without signing a redistribution schedule. Build the ledger against a delayed or sample tape first. Sit with it for a week. If the thing you actually wanted was to practice an idea, you may not need the live feed yet, and you certainly do not need an order route.

Stock Picks is the paper client we learned this on. Open it before you ask anyone, including us, for a price. If what you want is already there, the sprint you need might be a thin integration, not a platform. If what you want is custody of customer funds, the software line on the estimate is the cheap line, and we will say so before kickoff.